Matrix factorization to time-frequency distribution for structural health monitoring

Structural health monitoring enables structural information to be acquired through sensing technology, and is of need to early detect problems and damages in structures. Health monitoring strategies are often realized through a combination of qualitative sensing systems and high-performance structural integrity assessment methods. Structural deviations can be then effectively identified by interpreting the raw sensor measurements using signal processing techniques. The objective of this study is to develop a new structural health monitoring method that applies a matrix factorization algorithm to a time-frequency representation of multi-channel signals measured from a structure. This method processes vibrational input and/or output responses of structures to improve raw data quality, to estimate structural responses, to derive signal features, and to detect structural variations. For example, the proposed method can reduce the signal noise by utilizing first few principle vectors to reconstruct the measured signals. For frequency-domain responses, this method can smooth the phase to obtain a better input-output relationship of a structure. Additionally, the method removes abnormal signals in time series, allowing better understanding of structural behavior. Due to communication loss, this method is able to recover lost data from other channel measurements in a structure. Moreover, the proposed method transforms the signal components into a specific domain and then yield meaningful characteristics. All these features are numerically verified using experimental data, and the proposed method permits more detailed investigation of structural behavior.